[PATCH v2 08/12] riscv: dts: starfive: Add pool for coherent DMA memory on JH7100 boards

[Andrew Lunn]

On Sun, Oct 29, 2023 at 06:27:08AM +0200, Cristian Ciocaltea wrote:
> From: Emil Renner Berthing <emil.renner.berthing at canonical.com>
> 
> The StarFive JH7100 SoC has non-coherent device DMAs, but most drivers
> expect to be able to allocate coherent memory for DMA descriptors and
> such. However on the JH7100 DDR memory appears twice in the physical
> memory map, once cached and once uncached:
> 
>   0x00_8000_0000 - 0x08_7fff_ffff : Off chip DDR memory, cached
>   0x10_0000_0000 - 0x17_ffff_ffff : Off chip DDR memory, uncached
> 
> To use this uncached region we create a global DMA memory pool there and
> reserve the corresponding area in the cached region.
> 
> However the uncached region is fully above the 32bit address limit, so add
> a dma-ranges map so the DMA address used for peripherals is still in the
> regular cached region below the limit.
> 
> Link: https://github.com/starfive-tech/JH7100_Docs/blob/main/JH7100%20Data%20Sheet%20V01.01.04-EN%20(4-21-2021).pdf
> Signed-off-by: Emil Renner Berthing <emil.renner.berthing at canonical.com>
> Signed-off-by: Cristian Ciocaltea <cristian.ciocaltea at collabora.com>
> ---
>  .../boot/dts/starfive/jh7100-common.dtsi      | 24 +++++++++++++++++++
>  1 file changed, 24 insertions(+)
> 
> diff --git a/arch/riscv/boot/dts/starfive/jh7100-common.dtsi b/arch/riscv/boot/dts/starfive/jh7100-common.dtsi
> index b93ce351a90f..504c73f01f14 100644
> --- a/arch/riscv/boot/dts/starfive/jh7100-common.dtsi
> +++ b/arch/riscv/boot/dts/starfive/jh7100-common.dtsi
> @@ -39,6 +39,30 @@ led-ack {
>  			label = "ack";
>  		};
>  	};
> +
> +	reserved-memory {
> +		#address-cells = <2>;
> +		#size-cells = <2>;
> +		ranges;
> +
> +		dma-reserved {
> +			reg = <0x0 0xfa000000 0x0 0x1000000>;

If i'm reading this correctly, this is at the top of the first 4G of
RAM. But this is jh7100-common.dtsi. Is it guaranteed that all boards
derived from this have at least 4G? What happens is a board only has
2G?

It might also be worth putting a comment here about the memory being
mapped twice. In the ARM world that would be illegal, so its maybe not
seen that often. Yes, the commit message explains that, but when i
look at the code on its own, it is less obvious.

> +			no-map;
> +		};
> +
> +		linux,dma {
> +			compatible = "shared-dma-pool";
> +			reg = <0x10 0x7a000000 0x0 0x1000000>;
> +			no-map;
> +			linux,dma-default;
> +		};
> +	};